Using recent results in the measurement error literature, we show that the official U.S. unemployment rates substantially underestimate the true levels of unemployment, due to misclassification errors in labor force status in Current Population Surveys. Our closed-form identification of the misclassification probabilities relies on the key assumptions that the misreporting behaviors only depend on the true values and that the true labor force status dynamics satisfy a Markov-type property. During the period of 1996 to 2009, the corrected monthly unemployment rates are 1 to 4.6 percentage points (25% to 45%) higher than the official rates, and are more sensitive to changes in business cycles. Labor force participation rates, however, are not affected by this correction. We also provide results for various subgroups of the U.S. population defined by gender, race and age.